Bayesian logistic regression approaches to predict incorrect DRG assignment

Area of Science:

• Health Informatics
• Statistical Modeling
• Healthcare Management

Background:

• Diagnosis-Related Groups (DRGs) are crucial for inpatient care funding.
• Accurate DRG coding is essential for healthcare providers to ensure correct reimbursement.
• Clinical coding audits are resource-intensive, necessitating efficiency improvements.

Purpose of the Study:

• To implement and compare Bayesian logistic regression models for estimating DRG error probability.
• To assess the efficiency and accuracy of Bayesian approaches against classical methods in clinical coding audits.
• To identify factors influencing the likelihood of DRG errors.

Main Methods:

• Utilized Bayesian logistic regression models with weakly informative prior distributions.
• Estimated the probability of DRG revision for inpatient care episodes.
• Compared Bayesian models against each other and against maximum likelihood estimates.

Main Results:

• Bayesian models demonstrated superior parameter stability compared to maximum likelihood estimates.
• The best Bayesian model improved classification performance by 6% over maximum likelihood.
• Original DRG, coder, and coding day significantly impacted DRG error likelihood.

Conclusions:

• Bayesian approaches enhance model parameter stability and classification accuracy in DRG coding audits.
• The developed method offers improved operational efficiency for clinical coding audits.
• This statistical approach aids in optimizing healthcare resource allocation and funding accuracy.